On Classifying Latent Zeros Using Zero Inflated Models

• Activity Number: 655
• Type: Contributed
• Date/Time: Thursday, August 4, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Alok Kumar Dwivedi*+ and Rakesh Shukla and Marepalli B. Rao and Sada Nand Dwivedi and S.V. S. Deo
• Companies: University of Cincinnati and University of Cincinnati and University of Cincinnati and University of Dammam and All India Institute of Medical Sciences
• Keywords: Count data ; Zero inflated ; Nodal Involvement ; Latent zeros ; Negative binomial ; Classification

Abstract

Count data often has excess zeros in many clinical studies. These zeros can be a combination of latent zeros (low risk or high risk ) in some disease processes. Low risk zeros can arise due to absence of risk factors for disease initiation/progression and/or due to early detection of disease . High risk zeros can arise due to the presence of significant risk factors for disease initiation, progression or due to misclassification. Thus, we use latent class count model (zero inflated models) to estimate latent zeros using two separate processes and propose a strategy to further classify them. We use data on the number of involved nodes in breast cancer patients. We use predicted probabilities and posterior probabilities to classify patients into two latent classes of low and high risk negative nodes. Of 1152 patients studied, 38.8% were node- negative (zeros). The model predicted that 11.4% negative nodes are "high risk" and the remaining 27.4% are at "low risk" of nodal positivity. Posterior probabilities based classification was more appropriate compared to other methods. Our approach indicates that some patients with negative nodes may be re-assessed for their diagnosis about